Maintenance and disassembly method of large hydroelectric generating set based on deep learning algorithm

Abstract

The invention discloses a deep learning algorithm-based hydroelectric generating set maintenance disassembly planning method, and aims to solve the problems of high difficulty, low efficiency, resource waste and the like of large-scale hydroelectric generating set maintenance work. According to an original drawing, set equipment is subjected to three-dimensional modeling through SolidWorks software, and a three-dimensional visual model library containing equipment, maintenance tools and instruments and meters is constructed; then, determining disassembly information of the equipment to be disassembled, wherein the disassembly information comprises a disassembly priority sequence, a required tool and disassembly time; establishing a geometric constraint relationship between the equipment parts through the directed constraint graph and the constraint matrix; secondly, a disassembling sequence evaluation function is constructed by integrating quantitative indexes such as disassembling direction change, tool replacement, space movement distance and crane use conditions; finally, a deep learning algorithm is used for automatically optimizing and solving an optimal maintenance disassembly sequence; the maintenance efficiency of the hydroelectric generating set is effectively improved, time loss and resource waste are reduced, and the operation safety and efficiency of the hydroelectric generating set are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital maintenance and disassembly of large hydropower units, and in particular to a maintenance and disassembly method of large hydropower units based on a deep learning algorithm. Background Art

[0002] As a renewable energy generation method that uses the kinetic energy or potential energy of water to drive turbines and then convert them into electrical energy, hydropower is clean and stable, and is of great significance for ensuring energy supply and promoting green and low-carbon development. However, there are many technical challenges in the operation and maintenance of large hydropower units.

[0003] Large hydropower units have complex equipment structures and a large number of parts of various types, which makes their maintenance work particularly cumbersome. Since the units are usually installed inside hydropower stations, the space is relatively small, which further increases the difficulty of maintenance. In addition, the maintenance cycle of hydropower units is long, and various maintenance tasks need to be completed efficiently under the premise of ensuring safety.

[0004] At present, the maintenance space layout method based on two-dimensional drawings is still widely used. This method mainly relies on the experience of designers for planning, and has many shortcomings. First, the planning cycle is long and the cost is high, which makes it difficult to adapt to the rapidly changing maintenance needs. Secondly, the accuracy is poor, which easily leads to waste of resources, such as unnecessary replacement of parts or excessive maintenance. Finally, since two-dimensional drawings cannot fully reflect the three-dimensional structure and spatial relationship of the equipment, the spatial layout and disassembly sequence planning during the maintenance process lack scientific basis, which further increases the complexity and risk of maintenance.

[0005] Although with the development of technology, some advanced maintenance methods have gradually been introduced into the maintenance of hydropower units, such as virtual reality technology (VR) and augmented reality technology (AR), most of these technologies are still in the exploration and trial stage, and have not yet formed a complete and mature solution. In particular, for the disassembly sequence planning problem of large hydropower units, the existing technology still mainly relies on manual experience and trial and error, and lacks scientific and efficient planning methods and optimization tools.

[0006] Therefore, there is an urgent need for a maintenance and disassembly planning method for hydropower units that can fully consider the complexity of equipment structure, the diversity of parts, and the narrowness of maintenance space. This method should be able to automatically optimize and solve the optimal maintenance and disassembly sequence, improve the efficiency and quality of maintenance work, reduce time loss and resource waste, and thus ensure the normal operation and power generation efficiency of hydropower units.

[0007] In summary, the present invention aims to overcome the shortcomings of the prior art and provide a hydropower unit maintenance and disassembly planning method based on a deep learning algorithm to achieve the optimal design of the disassembly sequence of the hydropower unit, improve the efficiency and accuracy of the maintenance work, and provide strong support for the maintenance of the hydropower unit. Summary of the invention

[0008] The technical problem to be solved by the present invention is to provide a maintenance and disassembly method for large-scale hydropower units based on deep learning algorithms to solve the complexity and inefficiency problems faced by large-scale hydropower units in the maintenance and disassembly process. Specifically, hydropower generation is a clean, stable and renewable energy source. Its large-scale hydropower units have complex equipment structures, a wide variety of parts and components and a large number, long maintenance cycles, and usually need to be carried out in a small space. These problems not only increase the difficulty of maintenance work, but may also affect the normal operation and power generation efficiency of the hydropower units, thereby posing a serious threat to the safety and economy of hydropower generation. At present, the traditional maintenance space layout method based on two-dimensional drawings mainly relies on the experience of designers, and has problems such as long planning cycle, high cost, poor accuracy and waste of resources.

[0009] In order to solve the above technical problems, the technical solution adopted by the present invention is: a hydropower unit maintenance and disassembly planning method based on a deep learning algorithm, comprising the following steps: Step 1: Create a 3D visualization model for the maintenance of hydropower equipment; Step 2: Determine the dismantling information of the hydropower unit equipment to be dismantled; Step 3: Establish geometric constraint relationships between equipment parts through directed graphs and adjacency matrices, and express these constraint relationships through mathematical matrices. Then, the mathematical matrices are converted into relational tables and stored in the database. Step 4: Construct the encoding strategy and decoding strategy based on the disassembly constraints in Step 3, determine the quantitative indicators for evaluating the pros and cons of different disassembly processes, and then integrate the quantitative evaluation indicators to establish a mathematical model for the minimum disassembly time sequence planning; Step 5: Use the deep learning optimization algorithm to automatically solve the optimal or approximately optimal disassembly sequence based on the constraint relationship data and quantitative evaluation index data obtained in Step 3 and Step 4.

[0010] In a preferred solution, the three-dimensional visualization model in Step 1 includes three-dimensional models of equipment, maintenance tools and instruments, and the three-dimensional visualization model is a geometric model established in a 1:1 ratio using the software SolidWorks.

[0011] In a preferred solution, the disassembly information in Step 2 includes the priority of disassembly, the required disassembly tools and the estimated disassembly time.

[0012] In a preferred solution, the constraint relationship in Step 3 is established through a directed graph, and the mathematical matrix representation of the constraint relationship is: (1) In the formula, is the number of parts included in the hydropower unit equipment, For parts Whether to directly restrict components disassembly.

[0013] In a preferred solution, the quantitative indicators in Step 4 include the time variation caused by the change of disassembly direction, the time variation caused by the replacement of tools, the time variation caused by the moving distance of the maintenance personnel, and the time variation caused by the use of the bridge crane. In the preferred solution, the encoding strategy and decoding strategy in Step 4 include: giving priority to tasks with priority constraints and their immediate predecessor tasks when performing encoding tasks; the irreversibility of the disassembly process, that is, there is no repeated disassembly of each component in the system; the disassembly process has a unique initial state and end state.

[0014] In the preferred solution, the mathematical model of the disassembly sequence planning in Step 4 is: (2) In the formula, Represents a slave node To Node The difference in time required for the workload, , , , Respectively represent the nodes in the disassembly process To Node The time variation caused by the change of disassembly direction, the time variation caused by the replacement of tools, the time variation caused by the movement distance of the maintenance personnel, and the time variation caused by the use of the bridge crane; , , , Represent their respective weights.

[0015] In a preferred solution, the step of establishing a mathematical model for disassembling sequence planning in Step 4 includes: Step 4.1: Establish the cost of direction change , establish a three-dimensional rectangular coordinate system with the center of the disassembled equipment, then the disassembly direction of the parts can be used To define, then quantify the change in its direction, that is, the angle changes by 90°, the corresponding difference in workload is 1, and the change by 180°, the corresponding difference in workload is 2; Step 4.2: Create tools to change costs , Indicates that the node is disassembled To the disassembly node The difference in workload caused by tool changes. If the maintenance tools used by the two nodes are different, the value is 1, otherwise, it is 0; Step 4.3: Establish the change of personnel movement distance , Indicates that the node is disassembled To the disassembly node The difference in workload caused by the position change is calculated by the Euler distance between the two disassembled nodes and the corresponding models. To Node The calculation formula of the position change difference workload is as follows: (3) In the formula, is the measurement coefficient, The value of can be determined according to the distribution of specific equipment parts. To disassemble the node The x-axis coordinate of To disassemble the node The x-axis coordinate of To disassemble the node The y-axis coordinate of To disassemble the node The y-axis coordinate of To disassemble the node The z-axis coordinate of To disassemble the node The z-axis coordinate of Step 4.4: Establish the cost of using the bridge crane , Indicates that the node is disassembled To the disassembly node The difference in workload caused by the change in crane usage is 1 if both nodes use crane machinery, otherwise it is 0.

[0016] In a preferred embodiment, the steps of the deep learning optimization algorithm in Step 5 include input data transmission, hidden layer transmission, output layer calculation, error calculation, error back propagation and parameter update.

[0017] In a preferred solution, the specific steps of automatically solving the optimal or approximately optimal disassembly sequence based on the constraint relationship data and quantitative evaluation index data obtained in Step 3 and Step 4 in Step 5 include: Step 5.1: Input data transmission: the input data is passed to the first hidden layer through the input layer; Step 5.2: Hidden layer transfer, each hidden layer calculates the weighted sum according to the output of the previous layer, and generates output through the activation function; Step 5.3: Output layer calculation, the output of the last hidden layer is passed to the output layer for the final output calculation: (4) In the formula, For the final output, yes The weight of is the bias term, is the activation function; Step 5.4: Error calculation, calculate the error between the model output and the actual label; Step 5.5: Error back propagation, propagate the error back through the network, and calculate the gradient of each layer according to the chain rule; Step 5.6: Parameter update, according to the gradient descent rule, adjust the network parameters to reduce the error.

[0018] The maintenance and disassembly method of a large hydropower unit based on a deep learning algorithm provided by the present invention has the following beneficial effects: 1. The present invention optimizes and intelligently manages the maintenance and disassembly process of large hydropower units through three-dimensional visual modeling, determining disassembly information, establishing geometric constraint relationships, constructing disassembly sequence evaluation functions, and using deep learning algorithms to solve the optimal disassembly sequence, thereby solving the complexity and inefficiency problems faced by large hydropower units during the maintenance and disassembly process. 2. The three-dimensional visual modeling of the present invention: using professional software such as SolidWorks to perform accurate three-dimensional modeling, which improves the accuracy and practicality of the model; 3. The quantitative disassembly constraints of the present invention establish geometric constraint relationships between equipment parts through directed graphs and adjacency matrices, and realize the mathematization and storage of these constraint relationships; 4. The multi-index disassembly sequence evaluation of the present invention comprehensively considers multiple indicators such as direction change, tool replacement, spatial movement distance and crane usage, and constructs a comprehensive disassembly sequence evaluation function, which avoids unnecessary disassembly operations and time waste. The optimized disassembly sequence helps to reduce human operation errors and improve the safety and efficiency of the operation of the hydropower unit; 5. The application of the deep learning algorithm of the present invention introduces the deep learning algorithm to optimize the solution of the disassembly sequence, which improves the efficiency and accuracy of the calculation and significantly improves the efficiency of the maintenance work; 6. The precise disassembly sequence planning of the present invention helps to achieve optimal allocation of resources and reduce maintenance costs; 7. The present invention uses directed graphs and adjacency matrices to establish geometric constraint relationships between equipment parts, which can more accurately describe the constraint conditions during the disassembly process; 8. The present invention constructs a disassembly sequence evaluation function that comprehensively considers multiple indicators, which can more comprehensively evaluate the pros and cons of the disassembly sequence; 9. The present invention uses a deep learning algorithm to automatically solve the optimal or approximately optimal disassembly sequence, thereby improving the efficiency and accuracy of the calculation; 10. The present invention significantly improves maintenance efficiency and safety and reduces operation and maintenance costs through innovative disassembly technology and intelligent management system, providing solid technical support for the long-term stable operation of hydropower stations, and has higher application value and competitiveness in the field of maintenance and disassembly of large hydropower units. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present invention will be further described below in conjunction with the accompanying drawings and implementation examples: Figure 1 It is a schematic diagram of the workflow of the present invention; Figure 2 It is a directed graph of the hydropower station equipment to be disassembled in Example 2 of the present invention; Figure 3 This is a flow chart of a deep learning algorithm according to Embodiment 2 of the present invention; Figure 4 This is the maintenance and disassembly optimization diagram of the hydropower unit according to Example 2 of the present invention. DETAILED DESCRIPTION

[0020] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments: Example 1 like Figure 1 As shown, the maintenance and disassembly planning method of a hydropower unit based on a deep learning algorithm includes the following steps: Step 1: Create a 3D visualization model for the maintenance of hydropower equipment; Step 2: Determine the dismantling information of the hydropower unit equipment to be dismantled; Step 3: Establish geometric constraint relationships between equipment parts through directed graphs and adjacency matrices, and express these constraint relationships through mathematical matrices. Then, the mathematical matrices are converted into relational tables and stored in the database. Step 4: Construct the encoding strategy and decoding strategy based on the disassembly constraints in Step 3, determine the quantitative indicators for evaluating the pros and cons of different disassembly processes, and then integrate the quantitative evaluation indicators to establish a mathematical model for the minimum disassembly time sequence planning; Step 5: Use the deep learning optimization algorithm to automatically solve the optimal or approximately optimal disassembly sequence based on the constraint relationship data and quantitative evaluation index data obtained in Step 3 and Step 4.

[0021] In this embodiment, the three-dimensional visualization model in Step 1 includes three-dimensional models of equipment, maintenance tools and instruments, and the three-dimensional visualization model is a geometric model established in a 1:1 ratio using the software SolidWorks.

[0022] Furthermore, the disassembly information in Step 2 includes the priority of disassembly, the required disassembly tools and the estimated disassembly time.

[0023] Furthermore, the constraint relationship in Step 3 is established through a directed graph, and the mathematical matrix representation of the constraint relationship is: (1) In the formula, is the number of parts included in the hydropower unit equipment, For parts Whether to directly restrict components disassembly.

[0024] Furthermore, the quantitative indicators in Step 4 include the time variation caused by the change of disassembly direction, the time variation caused by the replacement of tools, the time variation caused by the moving distance of the maintenance personnel, and the time variation caused by the use of the bridge crane. Furthermore, the encoding strategy and decoding strategy in Step 4 include giving priority to tasks with priority constraints and their immediate predecessor tasks when performing encoding tasks; the irreversibility of the disassembly process, that is, there is no repeated disassembly of each component in the system; and the disassembly process has a unique initial state and end state.

[0025] Furthermore, the mathematical model of the disassembly sequence planning in Step 4 is: (2) In the formula, Represents a slave node To Node The difference in time required for the workload, , , , Respectively represent the nodes in the disassembly process To Node The time variation caused by the change of disassembly direction, the time variation caused by the replacement of tools, the time variation caused by the movement distance of the maintenance personnel, and the time variation caused by the use of the bridge crane; , , , Represent their respective weights.

[0026] Furthermore, the steps of establishing the mathematical model of the disassembly sequence planning in Step 4 include: Step 4.1: Establish the cost of direction change , establish a three-dimensional rectangular coordinate system with the center of the disassembled equipment, then the disassembly direction of the parts can be used To define, then quantify the change in its direction, that is, the angle changes by 90°, the corresponding difference in workload is 1, and the change by 180°, the corresponding difference in workload is 2; Step 4.2: Create tools to change costs , Indicates that the node is disassembled To the disassembly node The difference in workload caused by tool changes. If the maintenance tools used by the two nodes are different, the value is 1, otherwise, it is 0; Step 4.3: Establish the change of personnel movement distance , Indicates that the node is disassembled To the disassembly node The difference in workload caused by the position change is calculated by the Euler distance between the two disassembled nodes and the corresponding models. To Node The calculation formula of the position change difference workload is as follows: (3) In the formula, is the measurement coefficient, The value of can be determined according to the distribution of specific equipment parts. To disassemble the node The x-axis coordinate of To disassemble the node The x-axis coordinate of To disassemble the node The y-axis coordinate of To disassemble the node The y-axis coordinate of To disassemble the node The z-axis coordinate of To disassemble the node The z-axis coordinate of Step 4.4: Establish the cost of using the bridge crane , Indicates that the node is disassembled To the disassembly node The difference in workload caused by the change in crane usage is 1 if both nodes use crane machinery, otherwise it is 0.

[0027] Furthermore, the steps of the deep learning optimization algorithm in Step 5 include input data transmission, hidden layer transmission, output layer calculation, error calculation, error back propagation and parameter update.

[0028] Furthermore, the specific steps of automatically solving the optimal or approximately optimal disassembly sequence based on the constraint relationship data and quantitative evaluation index data obtained in Step 3 and Step 4 in Step 5 include: Step 5.1: Input data transmission: the input data is passed to the first hidden layer through the input layer; Step 5.2: Hidden layer transfer, each hidden layer calculates the weighted sum according to the output of the previous layer, and generates output through the activation function; Step 5.3: Output layer calculation, the output of the last hidden layer is passed to the output layer for the final output calculation: (4) In the formula, For the final output, yes The weight of is the bias term, is the activation function; Step 5.4: Error calculation, calculate the error between the model output and the actual label; Step 5.5: Error back propagation, propagate the error back through the network, and calculate the gradient of each layer according to the chain rule; Step 5.6: Parameter update, according to the gradient descent rule, adjust the network parameters to reduce the error.

[0029] Example 2 In another preferred embodiment, based on the above embodiment 1, refer to Figures 1 to 4 , this embodiment describes in detail the complete process of the hydropower unit maintenance and disassembly planning method based on deep learning algorithm.

[0030] S1: 3D Visual Modeling First, based on the original drawings of the hydropower station equipment, the unit equipment was 3D modeled using SolidWorks professional modeling software. During the modeling process, it was ensured that all equipment, maintenance tools, and instruments were accurately constructed at a 1:1 ratio to form a complete and accurate 3D model library. This step provided basic data support for subsequent disassembly sequence planning.

[0031] S2: Determine disassembly information Next, the hydropower unit equipment to be disassembled is analyzed in detail to determine its disassembly information; this information includes the disassembly priority, required disassembly tools, and estimated disassembly time; the priority is determined based on the constraint relationship between equipment components to ensure that the equipment is not damaged or disassembly is difficult due to improper sequence during the disassembly process; at the same time, the corresponding disassembly tools are prepared according to the disassembly requirements, and the time required for each disassembly step is estimated to facilitate the subsequent optimization of the disassembly sequence.

[0032] S3: Establishing geometric constraints Then, the geometric constraint relationship between equipment parts is established using directed graphs and adjacency matrices. The disassembly order and constraint relationship between parts are intuitively represented by directed graphs. The direction of the arrow indicates the disassembly direction, that is, the parts before the arrow must be disassembled before the parts after the arrow. At the same time, the constraint relationship is converted into a mathematical matrix form to facilitate subsequent calculation and optimization. These constraint relationships include different types such as free nodes, leaf nodes, ordinary nodes, and combined nodes, ensuring that the mutual influence between parts can be fully considered during the disassembly process.

[0033] S4: Constructing the evaluation function of the disassembly sequence After determining the disassembly information and geometric constraint relationship, a disassembly sequence evaluation function is constructed; this function comprehensively considers four indicators, including disassembly direction change, tool replacement, maintenance personnel movement distance and bridge crane usage, to quantitatively evaluate each disassembly sequence; specifically, the disassembly direction change is quantified by the angle change in the three-dimensional rectangular coordinate system; tool replacement is determined by comparing whether the tools used in adjacent disassembly steps are the same; the maintenance personnel movement distance is obtained by calculating the Euler distance between adjacent disassembly nodes; the use of the bridge crane is determined by whether a crane is needed during the disassembly process; these indicators are weighted and summed to obtain the total evaluation cost of the disassembly sequence, and the lower the cost, the higher the disassembly efficiency.

[0034] S5: Using deep learning algorithms to solve the optimal disassembly sequence Finally, the deep learning algorithm is used to solve the optimal or approximately optimal disassembly sequence. The disassembly information, geometric constraint relationship and disassembly sequence evaluation function are used as input data and trained and optimized through the deep learning network. The deep learning network includes multiple layers such as input layer, hidden layer and output layer. Through layer-by-layer transmission and calculation, the optimal or approximately optimal disassembly sequence is finally output. During the training process, the network parameters are continuously adjusted to reduce the error between the output and the actual label until the convergence state is reached. In this embodiment, after multiple iterations of training, Figure 2As shown in the figure, the optimal disassembly sequence is 1→2→7→5→18→19→20→9→8→10→15→16→17→11. The disassembly work cost corresponding to this sequence is the lowest and can meet the actual disassembly requirements.

[0035] Example 3 In another preferred embodiment, based on the above-mentioned embodiment 1 and embodiment 2, the diverse requirements in different disassembly scenarios are further considered.

[0036] Variant 1: Selection of different disassembly tools In the actual disassembly process, you may encounter situations where you need to use different disassembly tools. In order to fully demonstrate the flexibility and practicality of the present invention, this variant embodiment considers the selection of multiple disassembly tools in the disassembly sequence planning process. By constructing a disassembly tool library and adding relevant indicators of tool selection to the disassembly sequence evaluation function, the deep learning algorithm can automatically select the optimal disassembly tool combination during the training process, thereby further improving the disassembly efficiency and accuracy.

[0037] Variant 2: Adjustment of disassembly order For certain specific equipment or disassembly requirements, the disassembly order may need to be adjusted to optimize the disassembly process; this variant embodiment introduces a mechanism for adjusting the disassembly order in the disassembly sequence planning process; by modifying the priority order in the disassembly information or the directed graph in the geometric constraint relationship, the disassembly order can be flexibly adjusted to adapt to different disassembly scenarios and requirements; this mechanism enables the present invention to be more widely applied to the maintenance and disassembly work of various types of hydropower unit equipment.

[0038] Through the implementation and verification of the above embodiments, it is demonstrated that the present invention achieves efficient planning of the maintenance and disassembly process of large-scale hydropower units through a series of technical solutions such as building a three-dimensional visualization model, determining disassembly information, establishing geometric constraint relationships, constructing encoding and decoding strategies and quantitative evaluation indicators, and using deep learning algorithms to optimize disassembly sequences. This method not only improves the efficiency of maintenance work, reduces time loss and waste of resources, but also improves the safety and efficiency of hydropower unit operation.

[0039] In the preferred solution, in this embodiment, the three-dimensional visualization model in Step 1 includes three-dimensional models of equipment, maintenance tools and instruments, and the three-dimensional visualization model is a geometric model established in a 1:1 ratio using the software SolidWorks; the above settings ensure that the model is completely consistent with the actual equipment size, facilitate accurate simulation maintenance and drills in a virtual environment, and improve the accuracy and practicality of training. At the same time, the compatibility of SolidWorks also facilitates subsequent modification and expansion of the model.

[0040] In the preferred solution, the disassembly information in Step 2 includes the disassembly priority, the required disassembly tools and the estimated disassembly time; the above settings ensure the efficiency and accuracy of the disassembly process; at the same time, the solution also lists in detail the classification, storage method and safety handling measures of the disassembled parts, which fully guarantees the safety and standardization of the disassembly operation.

[0041] In the preferred solution, the quantitative indicators in Step 4 include the time change caused by the change in disassembly direction, the time change caused by the replacement of tools, the time change caused by the movement distance of the maintenance personnel, and the time change caused by the use of a bridge crane; the above settings are intended to comprehensively analyze and optimize the time consumption in the maintenance process, and provide data support for subsequent efficiency improvements and cost savings by accurately measuring various variable factors, thereby ensuring the efficiency and accuracy of maintenance work.

[0042] In the preferred solution, the encoding strategy and decoding strategy in Step 4 include: giving priority to tasks with priority constraints and their immediate predecessors when performing encoding tasks; the irreversibility of the disassembly process, that is, there is no repeated disassembly of each component in the system; the disassembly process has a unique initial state and end state; the above settings ensure the efficiency and accuracy of the disassembly process. At the same time, the first-in-first-out principle is used to manage the task queue to reduce task waiting time; an intelligent algorithm is introduced to optimize the disassembly sequence, further improve disassembly efficiency, and reduce energy consumption.

[0043] In the preferred scheme, the steps of the deep learning optimization algorithm in Step 5 include input data transmission, hidden layer transmission, output layer calculation, error calculation, error back propagation and parameter update; the above settings are intended to continuously adjust model parameters through an iterative training process, minimize prediction errors, improve the accuracy and generalization ability of the deep learning model, and ensure that the model performs well when processing complex tasks.

[0044] In summary, the maintenance and disassembly method of large-scale hydropower units based on deep learning algorithm proposed in the present invention realizes the optimization and intelligent management of the maintenance and disassembly process of large-scale hydropower units through the steps of three-dimensional visual modeling, determining disassembly information, establishing geometric constraint relationships, constructing disassembly sequence evaluation functions, and solving the optimal disassembly sequence using deep learning algorithms. The present invention first proposes to use professional software such as SolidWorks to perform three-dimensional modeling of large-scale hydropower unit equipment at a 1:1 ratio to form a complete three-dimensional visual model library including equipment, maintenance tools and instruments. This approach breaks through the limitations of traditional two-dimensional drawings and provides a more intuitive and accurate reference for subsequent maintenance and disassembly work. Different from the previous maintenance and disassembly planning that relied on the experience of designers, the present invention innovatively introduces deep learning algorithms, and significantly improves the accuracy and efficiency of planning by automatically optimizing and solving the optimal maintenance and disassembly sequence. When constructing the disassembly sequence evaluation function, the present invention comprehensively considers multiple indicators such as direction change, tool replacement, spatial movement distance and crane usage, and assigns Different weights are given to them, so as to more comprehensively evaluate the advantages and disadvantages of different disassembly processes; the present invention uses directed graphs and adjacency matrices, and the scheme successfully converts the geometric constraint relationship between equipment parts into a mathematical matrix representation, and then converts it into a relational table form and stores it in the database, providing reliable data support for subsequent optimization algorithms; the deep learning optimization algorithm proposed in the present invention can automatically solve the optimal or approximately optimal disassembly sequence based on constraint relationship data and quantitative evaluation index data. This algorithm is groundbreaking in the field of maintenance and disassembly of large hydropower units; after solving the optimal disassembly sequence, the present invention also verifies its feasibility through simulation, and outputs it in the form of an overall layout optimization plan for the maintenance parts of the hydropower unit, providing powerful guidance for actual production operations; this approach not only improves the efficiency of maintenance work, but also reduces costs and improves the safety and efficiency of the operation of hydropower units; in addition, the invention also has strong scalability and adaptability, and can be customized according to the characteristics and needs of different hydropower units, opening up a new path for the intelligent and refined management of hydropower units.

Claims

1. A hydropower unit maintenance and disassembly planning method based on deep learning algorithm, characterized in that: The method comprises the following steps: Step 1: Create a 3D visualization model for the maintenance of hydropower equipment; Step 2: Determine the dismantling information of the hydropower unit equipment to be dismantled; Step 3: Establish geometric constraint relationships between equipment parts through directed graphs and adjacency matrices, and express these constraint relationships through mathematical matrices. Then, convert the mathematical matrices into relational tables and store them in the database. Step 4: Construct the encoding strategy and decoding strategy based on the disassembly constraints in Step 3, determine the quantitative indicators for evaluating the pros and cons of different disassembly processes, and then integrate the quantitative evaluation indicators to establish a mathematical model for the minimum disassembly time sequence planning; Step 5: Use the deep learning optimization algorithm to automatically solve the optimal or approximately optimal disassembly sequence based on the constraint relationship data and quantitative evaluation index data obtained in Step 3 and Step 4.

2. The method for planning maintenance and disassembly of a hydropower unit based on a deep learning algorithm according to claim 1 is characterized in that: The three-dimensional visualization model in Step 1 includes three-dimensional models of equipment, maintenance tools and instruments, and the three-dimensional visualization model is a geometric model established in a 1:1 ratio using the software SolidWorks.

3. The method for planning maintenance and disassembly of a hydropower unit based on a deep learning algorithm according to claim 1 is characterized in that: The disassembly information in Step 2 includes the disassembly priority, required disassembly tools and estimated disassembly time.

4. The method for planning maintenance and disassembly of a hydropower unit based on a deep learning algorithm according to claim 1 is characterized in that: The constraint relationship in Step 3 is established through a directed graph, and the mathematical matrix representation of the constraint relationship is: （1）； In the formula, is the number of parts included in the hydropower unit equipment, For parts Whether to directly restrict components disassembly.

5. The method for planning maintenance and disassembly of a hydropower unit based on a deep learning algorithm according to claim 1 is characterized in that: The quantitative indicators in Step 4 include the time variation caused by the change of disassembly direction, the time variation caused by the replacement of tools, the time variation caused by the moving distance of the maintenance personnel, and the time variation caused by the use of the bridge crane.

6. The method for planning maintenance and disassembly of a hydropower unit based on a deep learning algorithm according to claim 1 is characterized in that: The encoding strategy and decoding strategy in Step 4 include: giving priority to tasks with priority constraints and their immediate predecessor tasks when performing encoding tasks; the irreversibility of the disassembly process, that is, there is no repeated disassembly of each component in the system; and the disassembly process has a unique initial state and end state.

7. The method for planning maintenance and disassembly of a hydropower unit based on a deep learning algorithm according to claim 1 is characterized in that: The mathematical model of the disassembly sequence planning in Step 4 is: （2）； In the formula, Represents a slave node To Node The difference in time required for the workload, , , , Respectively represent the nodes in the disassembly process To Node The time variation caused by the change of disassembly direction, the time variation caused by the replacement of tools, the time variation caused by the movement distance of the maintenance personnel, and the time variation caused by the use of the bridge crane; , , , Represent their respective weights.

8. The method for planning maintenance and disassembly of a hydropower unit based on a deep learning algorithm according to claim 7 is characterized in that: The steps of establishing the mathematical model of the disassembly sequence planning in Step 4 include: Step 4.1: Establish the cost of direction change , establish a three-dimensional rectangular coordinate system with the center of the disassembled equipment, then the disassembly direction of the parts can be used To define, then quantify the change in its direction, that is, the angle changes by 90°, the corresponding difference in workload is 1, and the change by 180°, the corresponding difference in workload is 2; Step 4.2: Create tools to change costs , Indicates that the node is disassembled To the disassembly node The difference in workload caused by tool changes. If the maintenance tools used by the two nodes are different, the value is 1, otherwise, it is 0; Step 4.3: Establish the change of personnel movement distance , Indicates that the node is disassembled To the disassembly node The difference in workload caused by the position change is calculated by the Euler distance between the two disassembled nodes and the corresponding models. To Node The calculation formula of the position change difference workload is as follows: （3）； In the formula, is the measurement coefficient, The value of can be determined according to the distribution of specific equipment parts. To disassemble the node The x-axis coordinate of To disassemble the node The x-axis coordinate of To disassemble the node The y-axis coordinate of To disassemble the node The y-axis coordinate of To disassemble the node The z-axis coordinate of To disassemble the node The z-axis coordinate of Step 4.4: Establish the cost of using the bridge crane , Indicates that the node is disassembled To the disassembly node The difference in workload caused by the change in crane usage is 1 if both nodes use crane machinery, otherwise it is 0.

9. The method for planning maintenance and disassembly of a hydropower unit based on a deep learning algorithm according to claim 1, characterized in that: The steps of the deep learning optimization algorithm in Step 5 include input data transmission, hidden layer transmission, output layer calculation, error calculation, error back propagation and parameter update.

10. The method for planning maintenance and disassembly of a hydropower unit based on a deep learning algorithm according to claim 1, characterized in that: The specific steps of automatically solving the optimal or approximately optimal disassembly sequence based on the constraint relationship data and quantitative evaluation index data obtained in Step 3 and Step 4 in Step 5 include: Step 5.1: Input data transmission: the input data is passed to the first hidden layer through the input layer; Step 5.2: Hidden layer transfer, each hidden layer calculates the weighted sum according to the output of the previous layer, and generates output through the activation function; Step 5.3: Output layer calculation, the output of the last hidden layer is passed to the output layer for the final output calculation: （4）； In the formula, For the final output, yes The weight of is the bias term, is the activation function; Step 5.4: Error calculation, calculate the error between the model output and the actual label; Step 5.5: Error back propagation, propagate the error back through the network, and calculate the gradient of each layer according to the chain rule; Step 5.6: Parameter update, according to the gradient descent rule, adjust the network parameters to reduce the error.

Images